AI
AI Optimization Tightens the Fastest Known Matrix-Multiplication Bound
A new research preprint combines gradient-based optimization, GPU-scale parallelism, and AlphaEvolve to lower the best known upper bound…
Section
AI
Artificial Intelligence (AI)
From the newsroom
Stories
AI
Qwen3.8-27B Tests How Much AI Can Fit Into a Smaller Footprint
Alibaba’s new 27-billion-parameter open-weight model keeps its predecessor’s broad scale while claiming gains across coding, agent and computer-use…
AI
Humanoid Robots Need a Better Test Than Pose Accuracy
HumanTracker pairs 153 hours of motion data with a learned score designed to catch the contact and stability…
AI
GitHub Turns Pull Requests Into an AI Agent Control Plane
GitHub agent apps bring specialized analytics, security, rollout, and operations agents into issues and pull requests, shifting software…
AI
AI Can Write Verified Code—Until the Repository Fights Back
Two new studies show why repository-scale correctness, repeatable execution and system coordination remain harder than generating plausible code.
AI
AI Agents Are Developing an Instruction-Maintenance Problem
Three new studies show why agent instructions need provenance, contract-aware compression, and traceable reuse as they become operational…
AI
AI Agent Safety Is Moving Beyond the Model
Three new papers show why safer AI agents need provenance-aware rules, evolving harnesses, recovery paths, and protected validation…
AI
Turning Telecom Data Into Images Can Cut AI Energy Use
A new study finds that plotting telecom time-series can reduce AI input tokens and energy use, while independent…
AI
AI Models Need to Learn When Context Deserves Trust
A new 23-model study argues that reliable AI must reject misleading context without learning to ignore information that…